Project Goals

This project is studying the role of epigenetic regulation by microRNAs in diabetes mellitus (DM) during ischemia/reperfusion (I/R) injury. During I/R, mTOR is active and is associated with Acute Myocardial Infarction (AMI). The suppression of mTOR by Rapamycin (RAPA) during I/R provides cardioprotection. This study is looking at the role miRNAs play during I/R with and without RAPA treatment. The primary goal of this analysis is to perform a differential expression analysis of the miRNA array data followed by a target analysis to see which genes these miRNAs are targeting.

Data Set Description

MicroRNA microarrays were run by LCSciences on their \(\mu\)Paraflo technology with the target miRNA probes being a mixture of Rabbit and Rat miRNAs. A total of 3 rabbit heart samples were pooled for each of the 3 conditions (DM, DM+I/R, DM+I/R+RAPA). According to the LCSciences data provided to me, 3 microRNA chips were run, 1 for each condition. On each chip, there are 4 probes that target each miRNA, which were treated as technical replicates by LCSciences. A summary of this experimental design is below:

*

WARNING: Replicate probes on the same chip are always used to obtain a single mean signal intensity for that condition [1]. It is not statistically valid to perform a differential expression analysis using technical replicates. Generally, at least 3 biological replicates are required for this type of analysis.

[1] Tang X, Gal J, Zhuang X et al. A simple array platform for microRNA analysis and its application in mouse tissues. RNA 2007;13:1803–22.

LCSciences Analysis

LCSciences has already pre-processed the data, including background subtraction and normalization. The also did a “Standard” and “In-Depth” analysis of the array data. Using the normalized data, LCSciences calculated the mean signal intensity for each miRNA using the 4 duplicate probes on each array. They also calculated differential expression p-values by treating these duplicated probes as technical replicates, but did not calculate fold changes. Technically, these technical replicates are NOT supposed to be used for differential expression analysis per the LCSciences paper [2]! Quote,

“The use of technical replicates alone will not be able to assess the biological variation within the same groups, will underestimate the standard deviation in Eq. 1, and therefore lead to false-positive calls.”

Thus, the data show below, especially the p-values, should be interpreted loosly. If there is a miRNA of interest it should be followed up with additional experimentation.

[2] Zhou X, Zhu Q, Eicken C et al. MicroRNA Profiling Using µParaflo Microfluidic Array Technology. In: Fan J-B (ed.). Next-Generation MicroRNA Expression Profiling Technology: Methods and Protocols. Totowa, NJ: Humana Press, 2012, 153–82.

New Analysis

Preprocessing

The LCSciences team has already performed background subtraction and normalization on the data, so it is being used as-is without further preprocessing. Additionally, the differential expression p-values calculated by LCSciences from the technical replicates are also being used as-is because other tools assume biological replicates are being input so the results won’t be any more valid. It looks like a standard student’s t-test has been used to determine if there is a significant difference in expression across 2 conditions from the LCSciences analysis.

From the LCSciences “Data Summary_S170089.doc” report, probes were filtered using the recommendations in the report. Probes that have a CV greater than 0.5 in any one condition are removed. Additionally, any probe with a detection p-value greater than 0.01 in any one condition is also removed because it was not detected significantly over the background signal.

Issue: A major issue is that there are only 402 miRNAs included in the differential expression analysis by LCSciences while there are really over 700 probes. According to the manual these are supposed to be filtered on StDev and Detection P-value, but it doesn’t look like they did that, because when I did the filtering I only get about 188 probes.

Contrasts

In the following sections I will work through each contrast. The filtering is done on a per-contrast level initially; thus, a miRNA may appear in one contrast, but not in another. The final heatmap lists miRNAs that passed these filters for all 3 contrasts.

Keep in mind that the p-values are calculated from duplicate probes, so are not trustworthy.

DM vs DM+I/R

In this section we are comparing the DM (control) condition versus DM+I/R. The DM condition is the control, so positive fold changes mean a miRNA is up-regulated in the DM+I/R condition compared to DM. Results are as follows:

  • CV and Detection P-value Filter leaves 194 miRNAs.
  • Differential Expression using Duplicate Probes leaves: 37 miRNAs.
Table and Volcano plot of differentially expressed miRNAs for DM vs I/R:

Warning: p-values are calculated from duplicate probes, so are not trustworthy.

log2FC.DMvsDMIR padj.DMvsDMIR
rno-miR-29c-3p -2.2120764 0.0000066
rno-miR-29c-5p -2.2943921 0.0000161
rno-miR-127-3p 1.4323323 0.0000211
rno-miR-20b-5p -1.6548968 0.0000499
rno-miR-146a-5p 1.0004054 0.0000667
rno-miR-378b -2.0790792 0.0000977
rno-let-7f-5p 1.2441330 0.0001101
rno-miR-140-3p 1.2630344 0.0001183
rno-miR-152-3p 1.0355182 0.0001351
rno-miR-29b-5p -2.6849039 0.0003277
rno-miR-148a-3p 0.8686883 0.0003693
rno-miR-762 1.9989808 0.0003868
rno-miR-193a-3p -0.8120581 0.0003925
rno-miR-301a-3p -2.3512721 0.0004104
rno-miR-495 1.5277619 0.0004545
rno-miR-379-5p 1.4540445 0.0005399
rno-miR-652-3p -1.2428018 0.0005617
rno-miR-193b-3p -0.8509218 0.0006111
rno-miR-208a-5p -1.8559897 0.0006362
rno-miR-378a-3p -2.1804392 0.0007468
rno-miR-30c-5p -1.3315645 0.0009549
rno-miR-15b-5p 1.1882454 0.0009675
rno-miR-499-5p -2.4012921 0.0011180
rno-let-7e-5p 1.2310875 0.0014059
rno-miR-5132-5p 1.6603443 0.0031994
rno-miR-154-5p 1.1213063 0.0032840
rno-miR-30b-5p -1.2778736 0.0035622
rno-miR-378a-5p -2.4676449 0.0039801
rno-miR-3075 1.0060490 0.0040481
rno-let-7i-5p 0.9977849 0.0041442
rno-miR-25-3p 1.0029013 0.0046547
rno-miR-328a-5p 1.4984830 0.0051956
rno-let-7d-5p 0.8875487 0.0062133
rno-miR-331-3p -1.0207314 0.0062441
rno-miR-328a-3p -1.1490298 0.0073097
rno-miR-100-5p 1.0041411 0.0077166
rno-miR-194-5p -2.2074694 0.0096370
## Loading required package: ggplot2
## Loading required package: ggrepel

DM vs DM+I/R+RAPA

In this section we are comparing the DM (control) condition versus DM+I/R+RAPA. The DM condition is the control, so positive fold changes mean a miRNA is up-regulated in the DM+I/R+RAPA condition compared to DM. Results are as follows:

  • CV and Detection P-value Filter leaves 196 miRNAs.
  • Differential Expression using Duplicate Probes leaves: 53 miRNAs.
Table and Volcano plot of differentially expressed miRNAs for DM vs RAPA:

Warning: p-values are calculated from duplicate probes, so are not trustworthy.

log2FC.DMvsRAPA padj.DMvsRAPA
rno-miR-223-3p 3.2386378 0.0000005
rno-miR-378b -1.3634159 0.0000017
rno-miR-494-3p 2.3245429 0.0000034
rno-miR-328a-3p -1.7129307 0.0000546
rno-miR-5132-5p 2.2781557 0.0000556
rno-miR-324-3p -1.4457998 0.0000627
rno-miR-3574 0.3679755 0.0001268
rno-miR-29b-3p 1.3448010 0.0002867
rno-miR-29c-5p -1.7262753 0.0003203
rno-let-7d-3p 1.1321560 0.0004077
rno-miR-106b-3p -1.0418202 0.0004128
rno-miR-320-3p 0.9325738 0.0004821
rno-miR-195-5p -1.1729918 0.0005576
rno-let-7a-5p 1.7956176 0.0006303
rno-miR-485-3p 1.2223924 0.0007736
rno-miR-93-3p -1.5563933 0.0007744
rno-miR-188-5p 1.3955851 0.0009708
rno-miR-497-5p -2.0979752 0.0010331
rno-let-7c-5p 1.6643880 0.0010651
rno-miR-499-5p 0.6809247 0.0010745
rno-miR-532-5p -1.0972972 0.0011295
rno-miR-483-5p 3.2571234 0.0011928
rno-miR-214-3p 1.1425832 0.0011935
rno-miR-125a-5p 0.9953577 0.0012161
rno-miR-34c-3p 1.1429580 0.0013502
rno-miR-378a-3p -1.3763183 0.0016222
rno-miR-30e-3p 1.0532229 0.0018864
rno-miR-100-5p 0.9284467 0.0020453
rno-miR-29b-5p -1.6940639 0.0023406
rno-miR-106b-5p -1.3219281 0.0023788
rno-miR-342-3p 1.3656495 0.0024719
rno-miR-466b-2-3p 1.4277486 0.0026090
rno-miR-3584-5p 1.7988601 0.0026427
rno-miR-107-3p -1.0810755 0.0027753
rno-miR-762 2.6732242 0.0027783
rno-miR-1-3p 3.1872030 0.0040209
rno-miR-23a-3p 1.0145204 0.0040934
rno-let-7b-5p 1.2958454 0.0043913
rno-miR-9a-5p 0.9406215 0.0044829
rno-miR-194-5p -2.4036119 0.0048198
rno-miR-21-5p 3.0923681 0.0049603
rno-miR-1224 1.0105218 0.0059565
rno-miR-378a-5p -2.1167378 0.0062297
rno-miR-143-3p -1.1499172 0.0063417
rno-miR-29a-3p -0.5063726 0.0070461
rno-miR-3596a 1.0904550 0.0071991
rno-miR-24-1-5p -3.3630973 0.0075028
rno-miR-149-3p 3.6009810 0.0079882
rno-miR-466c-3p 1.2651628 0.0081289
rno-let-7f-5p 2.0401048 0.0083018
rno-miR-103-3p -1.0641901 0.0090545
rno-miR-350 1.5563933 0.0093755
rno-miR-352 3.9108925 0.0097557

DM+I/R vs DM+I/R+RAPA

In this section we are comparing the DM (control) condition versus DM+I/R+RAPA. The DM condition is the control, so positive fold changes mean a miRNA is up-regulated in the DM+I/R+RAPA condition compared to DM. Results are as follows:

  • CV and Detection P-value Filter leaves 243 miRNAs.
  • Differential Expression using Duplicate Probes leaves: 52 miRNAs.
Table and Volcano plot of differentially expressed miRNAs for DM vs RAPA:

Warning: p-values are calculated from duplicate probes, so are not trustworthy.

log2FC.DMIRvsRAPA padj.DMIRvsRAPA
rno-miR-672-5p -2.3185691 0.0000003
rno-miR-539-5p -1.7535168 0.0000004
rno-miR-136-3p -1.9068906 0.0000017
rno-miR-499-5p 3.0822168 0.0000056
rno-miR-29b-3p 3.7310182 0.0000105
rno-miR-494-3p 2.0895870 0.0000117
rno-miR-483-5p 1.4987445 0.0000163
rno-miR-133b-3p 1.6479251 0.0000206
rno-miR-133a-3p 1.4559167 0.0000528
rno-miR-652-5p 1.1565703 0.0000649
rno-miR-411-3p -1.8845228 0.0001169
rno-miR-146a-5p -1.1853093 0.0001499
rno-miR-667-5p 1.7900769 0.0001871
rno-miR-872-5p -1.0303245 0.0001930
rno-let-7d-3p 1.0746049 0.0002407
rno-miR-352 2.3943170 0.0003876
ocu-miR-498 0.4355016 0.0004281
rno-miR-148a-3p -0.7315695 0.0004717
rno-miR-152-3p -1.4330440 0.0005860
rno-miR-196c-3p 2.8729947 0.0011605
rno-miR-379-3p -2.0170735 0.0013315
rno-miR-374-5p 1.8110973 0.0015749
rno-miR-127-3p -1.8035268 0.0015808
rno-miR-339-5p 0.6069888 0.0016109
rno-miR-29c-3p 2.1351303 0.0019827
rno-miR-145-5p -0.8330058 0.0021377
rno-miR-144-5p -1.6455040 0.0021410
rno-miR-497-5p -2.4979058 0.0021830
rno-miR-15a-5p 1.3799040 0.0022085
rno-miR-34c-3p 1.1794838 0.0022557
rno-miR-485-3p 1.0435893 0.0022763
rno-miR-495 -1.9385995 0.0024422
rno-miR-7a-1-3p 1.1369356 0.0027040
rno-miR-30e-5p 1.3458661 0.0027079
rno-miR-143-3p -1.1047057 0.0027498
rno-miR-378a-3p 0.8041209 0.0035076
rno-miR-140-3p -1.7274615 0.0037601
rno-miR-543-3p -2.8237494 0.0041975
rno-miR-5132-5p 0.6178114 0.0042914
rno-miR-322-5p 2.9149579 0.0043069
rno-miR-30c-5p 1.0530019 0.0045680
rno-miR-206-3p -1.5008982 0.0051527
rno-miR-140-5p -2.2411454 0.0060138
rno-let-7a-5p 1.1434264 0.0060287
rno-miR-133a-5p 1.8121400 0.0074002
rno-miR-26b-5p 1.1297868 0.0074592
rno-miR-290 1.5181310 0.0078814
rno-miR-125a-3p 0.6709357 0.0081299
rno-miR-466b-5p -1.1856923 0.0082058
rno-miR-6215 1.1989810 0.0085748
rno-miR-378b 0.7156632 0.0090364
rno-miR-30b-5p 1.0656578 0.0097763

Heatmap Analysis

Below are two heatmaps of the data. The first shows the scaled signal intensities of each probe that passed the CV and Detection P-value filter for all 3 conditions. The second shows the fold changes for this same cohort of miRNAs.

The following is a heatmap of the log2 of the mean signal intensities for each condition. There are 188 miRNAs that passed the filteres for all 3 conditions (included in heatmaps below), with 69 miRNAs that were excluded because they failed in one or more conditions.

The following is a heatmap of the log2 Fold Change for each contrast.

## Warning: The input is a data frame, convert it to the matrix.